During earnings calls, C-suite executives provide context to the presented financial numbers and, therefore, help to build a narrative around a company’s financial performance. These sessions can also unveil future risks and opportunities the numbers haven’t reported. Hence, earnings calls and their accompanying transcripts are a powerful data resource to build a more comprehensive understanding of a company’s financial outlook, alongside public accounting data.
However, unlike public accounting or market data, which is structured, earnings call transcripts present unstructured data via words and phrases. This is where natural language processing (NLP) becomes crucial in extracting meaningful insights from earnings calls. NLP can help decode complex financial language, identify sentiment, and highlight key themes discussed during the call.
Historically, word count methods such as ‘Bag of Words’ have long been a widely used technique for analyzing text data. However, they have limitations and shortcomings. For example, they cannot extract information about the relationship between words within a document. By contrast, more modern NLP inference techniques are able to consider context by using textual data embeddings such as FinBERT or transformer-based deep learning algorithms such as GPT-3 or GPT-4.
This use of NLP techniques means we can infer the average sector sentiment during earnings call conferences. Our animation below tracks the swings in net sector sentiment identified during earnings call conferences of S&P 500 companies over time, starting in 2014.
Source: Robeco, FactSet. The animation shows the average net sentiment for the top five sentiment GIGS sectors in earnings call conferences over the last 10 years. For each company and quarter, the net sentiment is computed as the probability that the transcript text sentiment is positive minus the probability that its sentiment is negative. The analysis includes all S&P 500 constituents, and the sample period ends on August 9, 2024. Eighty-seven percent of the S&P 500 constituents already had their Q2 2024 earnings call conferences.
However, we also observe diverging sentiment across sectors. For instance, over the last year, the information technology, communications services, and healthcare sectors each experienced a material jump in sentiment scores, corresponding to positive developments in artificial intelligence, digital media, and glucagon-like peptide 1 (GLP-1) weight loss medications such as Ozempic. Conversely, while economic conditions have improved, the cumulative effect of inflation has weighed on sentiment for both the consumer discretionary and consumer staples sectors.
The analysis above highlights the application of NLP for dynamic sentiment detection using earnings calls. To explore how such tools might be used for dynamic quantitative theme investing, we invite you to contact your Robeco sales representative.
Holen Sie sich die neuesten Einblicke
Abonnieren Sie unseren Newsletter, um aktuelle Anlageinformationen und Analysen durch Sachverständige zu erhalten.
Active Quant: Mit Zuversicht Alpha anstreben
Alpha sollte aber mehr sein als eine Illusion. Wir lassen nichts unversucht, um für unsere Kunden Alpha zu erzielen.